You are on page 1of 20

Chapter 3


The VCO used in the traditional PLL frequency synthesizer is an external

block which is either a module or built with high performance discrete com-
ponents to meet the requirements with a fairly large margin. Therefore, the
synthesizer employing an external VCO decouples the VCO design from the
rest of the PLL design for phase noise consideration. Over designed for phase
noise, an external VCO offers sufficient degrees of freedom to meet the PLL
performance parameters independently. This is not the case for fully integrated
frequency synthesizer with the VCO in CMOS technology. The VCO phase
noise requirement can only be met with a small margin, and hence this strongly
couples PLL and VCO design for phase noise consideration.
This chapter explores the phase noise characteristics of a PLL frequency
synthesizer. First, the phase noise characteristics of the oscillator are investi-
gated with low phase noise design in mind. An analytical model is developed
to investigate the noise properties of the closed-loop PLL. The closed-loop PLL
noise is an important factor of the total system performance in modern digital
communications which use phase modulation, such as QPSK and QAM. The
PLL noise model includes the effect of the blocks forming the PLL as well as the
VCO. A SPICE implementation of the model is also compared with measured

1. Phase Noise Definition

The most critical specification for any oscillator is its spectral purity. The out-
put spectrum of an ideal oscillator is an impulse at, as shown in Fig. 3.1(a),
An ideal oscillator can be described as a pure sine wave in the time domain,

However‚ in any practical oscillator‚ the spectrum has power distributed

around the center frequency as shown in Fig. 3.1(b). In addition‚ the
power is also distributed at the harmonics of the oscillation frequency‚ i.e.‚
The instantaneous output of a practical oscillator may be
represented by;

where and represent amplitude and phase fluctuations of the signal‚ re-
spectively. There are two types of phase terms appearing at the output spectrum
of an oscillator. The first type appears as a distinct component in the spectrum‚
and it is referred to as a spurious tone or signal as shown in Fig. 3.1(b). The
second type appears as random phase fluctuations‚ and it is referred to as phase
noise as shown in Fig. 3.1(b). The phase noise in an oscillator is mainly due
to internal noise sources such as thermal noise and active device noise sources
( flicker noise or 1/f‚ shot noise). They are random in nature. The internal
noise sources set a fundamental limit for a minimum obtainable phase noise in
oscillator design.
The spurious tones are due to external noise sources such as noise on con-
trol voltage‚ power supply‚ and bias current coupled clock signals. They are
deterministic in nature. The spurious tones are not directly related to the oscil-
lator but are important in the frequency synthesizer output and the PLL design

The phase noise of an oscillator is typically quantified by the single-sideband

(SSB) phase noise‚ which is defined as the ratio of noise power in a 1 Hz
bandwidth at an offset‚ to the signal power‚ as shown in Fig. 3.2. Single-
sideband (SSB) phase noise is specified in at a given frequency offset‚
PLL Phase Noise Analysis 23

from the signal frequency.

Equation 3.3 is also can be written as‚

where is the rms value of the sinusoid representing the phase noise
sideband at the offset frequency and is the rms value of the carrier
Sidebands around the carrier shown in the Figure 3‚2 can be from PM or AM
modulation of the carrier signal by noise. The phase noise in Equation 3.3 is
written as AM and PM noise contributions as‚

where is the double-sideband (DSB) phase noise spectral density and

is the DSB amplitude noise spectral density.
Consider a signal of constant amplitude which is phase-modulated by a
sine wave of frequency

where is the peak phase deviation‚ also known as the modulation index.
When then the narrow band FM approximation can be used to

obtain [14]‚

The amplitude of the side band‚ due to modulating signal can be written

Also‚ from the above equation‚ the peak phase deviation‚ due to a single
sideband tone with amplitude‚ around the carrier is expressed as‚

Assuming that the side bands are due to the phase noise fluctuations‚ then the
phase noise power spectral density can be written using Equations 3.4 and 3.9


Equations 3.10 and 3.11 are of fundamental importance for the treatment‚ calcu-
lation and simulation of phase noise in PLLs. While is useful to measure
and characterize with measurement tools‚ is useful in calculations of in-
tegrated residual phase deviation over a given bandwidth [8].
The effect of AM modulation of a carrier signal is considered. An AM
modulated signal can be written as‚

where is the AM modulation index‚ is the modulation frequency. Equa-

tion 3.12 can be expanded as;

Equation 3.13 indicates that AM modulation generates a pair of spurious

sidebands similar to those of PM modulation (narrow-band). Only difference
between the sidebands of AM and PM is the phase relationship as shown in
Figure 3.3. When is measured by a spectrum analyzer‚ it does not
distinguish the AM and PM sidebands since phase information is not retained
by a spectrum analyzer. Close to the carrier‚ PM modulated noise dominates‚
and can be considered as phase noise. Far offset frequencies‚ both AM and PM
noise contribute equally [14].
PLL Phase Noise Analysis 25

2. Oscillator Noise Characteristics

Noise characteristics of oscillators have a great impact on the overall noise
characteristics of a PLL since each PLL typically employs two oscillators; a
reference crystal oscillator and an RF VCO. Phase noise of a crystal oscillator
has an impact on the close-in phase noise of a PLL while VCO noise shapes
phase noise of a PLL at large offset frequencies. Effects of individual noise
contribution of crystal and voltage controlled oscillators on the output PLL noise
will be discussed in the PLL noise analysis section. Phase noise characteristics‚
modeling and simulation of oscillators are overviewed in this section.
An oscillator noise profile exhibits different slopes at various regions when
the SSB phase noise is plotted in decibels versus log frequency as shown in
Figure 3.4. These regions have slopes of 0‚ (-10 dB/decade)‚ (-20
dB/decade) and (-30 dB/decade). The noise plot of an active device (MOS
or Bipolar transistor) is shown in Figure 3.5 which has only two regions; thermal
noise and region. The noise corner is in the vicinity of 500KHz to
1MHz for sub-micron CMOS technology [11]‚ and it is in the vicinity of 1KHz
to 10KHz for bipolar transistor.
The noise from the active and the passive devices are injected into the res-
onator of an oscillator. The injected noise is shaped by the VCO resonator which
has -20 dB/decade slope on either side of the oscillation frequency. The flat ther-
mal noise becomes in nature and the noise becomes in nature‚
within the resonator bandwidth‚ at the output of the oscillator (Figure 3.4(a)).
Depending on the relative position of the 1/f corner and the resonator band-
width‚ two cases arise as shown in Figure 3.4(a) and (b). Figure 3.4(a)
is typical characteristic of the most RF oscillators with LC resonator tanks.
Figure 3.4(b) is typical characteristic of a high-Q crystal oscillator.

Oscillator Noise Models and Simulation

Phase noise in oscillators has long been the subject of theoretical and experi-
mental investigation. An early model of phase noise‚ introduced by Leeson [19]‚

qualitatively described phase noise spectra in a variety of oscillators [20]. The

derivation of the Leeson phase noise model is based on a linear time-invariant
(LTI) approach to the analysis of noise in oscillators. The Leeson phase noise
model is given as‚

where is the loaded quality factor of the resonator‚ is the oscillation

frequency‚ is the signal power of the oscillator‚ is the frequency offset‚
F is the noise factor for the active devices‚ is the Boltzmann constant‚ T is
the temperature in Kelvin‚ and is the flicker noise corner frequency knee in
the phase noise. does not necessarily coincide with the flicker noise corner
frequency of the active device.
Equation 3.14 shows clearly that three parameters Q‚ F‚ and have signif-
icant effect on the phase noise of an oscillator. A low noise oscillator requires
minimum F and maximum Q and Minimization of the noise factor F in-
PLL Phase Noise Analysis 27

volves mostly active devices. MOS transistors have higher flicker noise corner
frequency (from 500KHz up-to a couple of MHz) than a typical bipolar tran-
sistor. Output power‚ should be maximized within a certain power budget.
The loaded Q of the LC-tank for fully integrated RF VCO is limited by the
available integrated spiral inductor and varactor in a given technology.
Phase noise models based on a linear time-invariant analysis provide insights
and quantitative understanding of noise in oscillators and serve as a starting point
in design procedure. An accurate estimation of the phase noise of an oscillator
requires nonlinear simulations including all available noise models.
Mathematically more rigorous solutions by Hajimiri [21] and Demir [22]
have been introduced in recent years since Leeson’s approximate model. How-
ever‚ these solutions are mathematically involved‚ do not provide much intuition
into design and analysis of circuits‚ and are not practical for hand calculations.
These solutions are more appropriate for simulator implementation and verifi-
SPICE small-signal noise simulation (.AC) does not provide adequate mod-
eling of an RF oscillator noise. The state of art RF simulators provide a steady-
state noise analysis which is based on the steady-state behavior of the circuit‚
rather than its DC operating point (EldoRF[23]‚ SpectreRF [24]‚ Agilent EEs of
[25]). The circuit is linearized around the large-signal periodic time-varying
operating point‚ and the noise contributions (the amplitudes of the noise sig-
nals) are assumed to be small enough so that the small-signal assumption is
valid. Close to carrier (offset frequencies phase noise simulations
may not give accurate results because small-signal assumptions are not valid
close to carrier (power level in the order of -30dBm and higher). However‚ it
is often not very critical to get very accurate result for the VCO phase noise
at very close offset frequencies for two reasons; first‚ most wireless standards
modulation schemes do not carry information at very close offset frequencies
to the carrier and second the VCO is placed in the PLL which suppresses the
VCO noise within the loop bandwidth close the carrier.

3. PLL Noise Analysis

Phase noise of a VCO placed inside a PLL is shaped by the PLL noise transfer
functions. A free running VCO phase noise is simply called VCO phase noise‚
while the phase noise of a VCO inside a locked PLL is called PLL output
phase noise. The overall PLL output phase noise is characterized by the noise
contributions of all blocks in a PLL.
A linear phase-domain model of a PLL with additive noise sources is shown in
Figure represents the noise in that appears at the reference
input to the PFD. It includes the crystal oscillator‚ crystal buffer‚ and reference
divider noises. is the noise due to the divider in is the
phase noise of the VCO in is the noise of PFD in

is the noise at the VCO control voltage input due to the loop filter and
other coupled noise sources to the control line in is the noise of
the CP current in The transfer functions from these noise sources to
the output can be written as;

The transfer functions, and are defined as (from Figure 3.6);

PLL Phase Noise Analysis 29

where is the CP current in, A. is the VCO gain in

is the loop filter transfer function.
The noise transfer functions are helpful to understand the effect of individual
block contributions to the PLL output noise. The total noise can be calculated
by adding all the block noise contributions in an RMS sum,


Analysis of the noise transfer function reveals more information about the
PLL noise shaping effect. The common factor for reference‚ divider‚ PFD and
CP noises is‚

The amplitude response of this common factor is shown in Figure 3.7(a). It

exhibits low pass characteristics‚ and at low frequencies‚ the noise of the PLL
is dominantly contributed by the reference‚ divider‚ PFD and CP noises. Also‚
the noise contributions from reference‚ divider‚ PFD and CP are increased by
the feedback divider ratio N inside the loop bandwidth.

The amplitude response of the common factor, for VCO

and the control line noises is shown in Figure 3.7(b) and, it exhibit high-pass
filter behavior. The VCO and the control line noises are suppressed inside the
loop bandwidth, and the VCO noise dominates beyond the loop bandwidth.
The PLL bandwidth determines the total PLL noise shape. For a minimum
residual phase error at the PLL output (integrated noise), the PLL loop band-
width is chosen at the intersection of close-in noise and VCO noise as shown
in Figure 3.8(a) [8]. Performance loss occurs if a non-optimum bandwidth is
used as shown in Figure 3.8(b) and (c).

A typical SSB phase noise of a PLL is shown in Figure 3.9 for a well designed
PLL. The phase noise behavior exhibits three different regions. The first phase
noise region is the reference oscillator noise (about 100Hz-1KHz). The second
region‚ also called PLL close-in noise‚ is dominantly contributed by PFD‚ CP‚
and the divider noise. The third region beyond the loop bandwidth is practically
the VCO noise. The first and third regions are primarily determined by the
oscillators; crystal and voltage-controlled. The second region‚ PLL close-in
noise‚ is the indicator of the performance of a given PLL.
From Figures 3.7(a) and 3.9‚ we can define a PLL noise floor‚

where is the PLL close-in phase noise in is the noise floor

of PLL (noise due to PLL circuitries; PFD‚ divider and CP). 20 log(N) is the
increase of phase noise from the frequency magnification due to the feedback
ratio‚ 1/N. is the increase of noise due to the reference frequency.
When the reference frequency value increases‚ the CP starts to inject more noise
into the loop filter. This will be further explained in the CP noise characteristics
PLL Phase Noise Analysis 31

For a measured PLL close-in noise‚ the noise floor of the PLL is found from‚

Equation 3.27 provides a figure of merit for the PLL Synthesizer circuit itself.
This allows comparison of different PLLs’ performance by accounting for the
feedback divide ratio‚ N and the reference frequency. Also‚ the noise floor of
a PLL can be used to quickly identify the performance of a given PLL once N
and the reference frequency are known.
Guidelines to optimize the output phase noise of a PLL frequency synthesizer
are listed below‚

To minimize close-in phase noise (noise inside the loop bandwidth of PLL);
reduce divider‚ CP‚ and PFD noises.

To minimize close-in phase noise (noise inside the loop bandwidth of PLL);
minimize divide ratio (N). The selection of divide ratio depends on the
frequency synthesizer architecture. The minimum divide ratio N is limited
by the output frequency resolution (channel bandwidth) in integer-N PLL
architectures. Fractional-N architectures may be used to minimize divide
ratio N [26].

To minimize close in phase noise (noise inside the loop bandwidth of PLL)
if dominated by CP noise; increase the CP current while keeping CP
noise at the same level.

To minimize phase noise outside the loop bandwidth of PLL; minimize VCO
noise‚ minimize noise on the control line and minimize VCO gain

PLL Noise Modeling and Simulation

The closed-loop noise of a PLL can be calculated by using the analytical
models developed in Section 3. Equation 3.23 can be implemented by using
MATLAB‚ MATHCAD‚ EXCELL or with similar software. An implementa-
tion of the linear PLL noise model in SPICE is developed for closed-loop PLL
noise simulation by representing phase variables as voltages and converting
phase noise into voltage noise. The SPICE model is chosen because of the
flexibility to include effects such as layout parasitics‚ bond-pads or active loop-
filter components. The developed linear model is shown in Figure 3.10. This
model includes the noise contribution of all PLL building-blocks. The CP and
loop filter noises are represented in voltage/current since they are treated as cur-
rent/voltage variable in the original loop. The reference‚ PFD‚ divider‚ prescaler‚
and VCO noises are converted from the SSB representation to phase
noise power spectral density using Equation 3.10.

The same model is also used to predict PLL stability and transient behavior.
The model allows the prediction of noise contribution from individual PLL
blocks‚ and hence it can be used for optimization of the overall PLL noise
performance. An optimized PLL noise performance leads to relaxed VCO
noise requirements by avoiding to overdesign.

Steps to Calculate Closed-loop Noise of a PLL

The major steps to calculate closed-loop noise of a PLL using the developed
model are;
PLL Phase Noise Analysis 33

Determine loop filter parameters for a given phase margin and loop band-
width along loop parameters (VCO gain‚ charge-pump gain‚ divide ratio‚
reference frequency).
Determine noise characteristics of individual PLL blocks by simulation or
Use the individual noise characteristics in the model to compute PLL output

Noise Characteristics of Individual PLL Blocks

Noise characteristics of individual PLL blocks must be extracted for PLL
noise simulation using the developed model. The noise characteristics of in-
dividual PLL blocks can be extracted through measurement or/and simulation.
The crystal oscillator phase noise characteristics under certain operation con-
ditions are given by manufactures. Also‚ it can be measured since it is off-chip.
The VCO noise can be characterized by measurement or simulation. The noise
characteristics of oscillators are discussed in Section 2.0.
However‚ it is usually difficult to extract phase noise information for the
other PLL blocks (divider‚ PFD‚ CP) from measurement. This is due to dif-
ficulty of a measurement setup for these blocks‚ and the noise levels of these
blocks are below most of the measurement instruments’ noise floors requiring
special expensive instruments. Simulation is more convenient to extract noise
information for these blocks.
Simulation of divider noise is straightforward. A pure single tone is ap-
plied at the divider input‚ and the SSB noise is measured at the divider output.
Figure 3.11 shows the simulation setup and the simulated phase noise of a
divide-by-8 prescaler.
A typical PFD and CP configuration is shown in Figure 3.12. The contribu-
tion of PFD noise is twice the noise of one flip-flop since both the flip-flops in
PFD contribute equally to the phase noise. The phase noise of one flip-flop in
PFD is simulated and used in the simulation model‚ and it is multiplied by 2 in
the model.
An ideal CP-PLL with zero phase error neither sources current to‚ nor sinks
current from‚ the loop filter. However‚ PLLs with zero phase error are insensitive
to small loop-phase deviations due to finite signal rise times in the PFD and
charge pump which is also called the “dead-zone” [17]. A commonly employed
solution to “dead-zone” problem is to use an artificial phase offset so that CP
pumps/sinks current when PLL is locked. When the PLL is locked‚ the average
output current flowing into the loop filter as shown in Figure 3.12 is zero. Both
the UP and DN currents are on for the duration of the “dead-zone” pulse. Even
though the average current is zero‚ noise is injected from both UP and the DN
currents for the duration of the “dead-zone” pulse. The charge pump current
PLL Phase Noise Analysis 35

noise injected to the loop filter under lock condition can be calculated as‚

where factor 2 is used to account for UP and DN current pulses. is

the current noise of the CP in is the dead-zone pulse width. T is
the period of the reference signal.
Equation 3.29 suggests that if the reference frequency is increased‚ the CP
will pump more noise into the loop filter. Hence‚ the PLL close-in noise given
by Equation 3.26 will increase with the reference frequency.
The practical implementation of charge-pump circuit presents non-ideal ef-
fects; (i) the leakage currents (ii) the mismatch between magnitudes of up
and down currents (iii) switching time mismatches between the up and down
pulses [17]. These non-ideal effects are important for the reference spur calcu-
lation. They do not have significant effect on the noise calculation presented
here if the CP circuit is properly designed. Equation 3.29 is sufficient for the
CP noise calculation. Two key parameters to reduce the CP noise contribution
to the PLL output noise are the dead-zone pulse width and the CP current noise.
The dead-zone pulse width and the CP current noise should be minimized to
minimize the CP noise contribution.

Experimental Results and Noise Optimization Techniques

Several PLL circuits have been analyzed using the developed SPICE model.
Simulated and measured results for a 4GHz PLL [27] are presented. The 4GHz
PLL is designed for 2.4GHz/5GHz multi-band WLAN radio applications. A
prototype is designed as a test structure. The 4GHz PLL is fully integrated
including loop filter components in CMOS technology. Only required
external sources are a reference clock signal, 1.8V supply voltage, and bias
currents for individual blocks.
The designed PLL loop parameters are N = 192,
PM = 53°, and The corresponding loop filter
parameters are and The open-loop
behavior is simulated by using the developed model by opening loop at the
CP output. The open-loop behavior is shown in Figure 3.13. However, the
measured PLL open-loop behavior is different than the inital design because
the measured VCO gain is higher than the inital design. The measured VCO
gain is about The open-loop behavior is re-simulated with the
measured VCO gain as shown in Figure 3.14. The loop bandwidth is moved to
and the phase margin is reduced to 39°. This change is verified with
measurement as shown in Figure 3.15. The measured loop bandwidth is about
The peaking at the loop bandwidth occurs due to reduced phase

margin. Figure 3.15 also shows both the simulated phase noise and measured
phase noises.
The closed-loop noise is simulated using measured phase noise for VCO and
external reference signal, and simulated noise for other blocks in the developed
SPICE model. The individual noise contributions at the PLL output from the
PLL blocks are shown in Figure 3.16. The close-in noise of PLL is dominated
by the CP noise. The close-in phase noise should improve by increasing the CP
current as suggested by the analytical models since the CP noise is dominating
close-in PLL noise. This is confirmed with further testing. Figure 3.17 shows
the measured and simulated PLL output noises for the CP current,
The close-in PLL noise improves as expected. Also, more peaking around
the loop bandwidth is observed for the CP current, This is
because of the further reduction of the phase margin. The open-loop behavior
for is shown in Figure 3.18. The loop bandwidth is moved
to and the phase margin is reduced to 24°. The measured loop
bandwidth in Figure 3.17 confirms the open-loop simulation.
The developed model is able to predict the total PLL noise within in this
example. An important conclusion from this example is that the fully integrated
loop filter components do not allow the optimization and evaluation of a PLL
performance. Optimization of PLL close-in noise involves iteration of CP
current value for minimum noise while keeping the loop characteristics constant.
This requires adjustment of the loop filter component values when the CP current
value is changed. Also, off-chip loop filter components allows characterization
of the VCO noise from measurement with a locked PLL by setting the loop
bandwidth very narrow Variation in loop parameters has significant
impact in the PLL performance.
PLL Phase Noise Analysis 37
PLL Phase Noise Analysis 39

4. Summary
The analysis‚ calculation and optimization of the PLL output noise are pre-
sented in this chapter. Oscillator noise characteristics have important impact
on the PLL phase noise since each PLL frequency synthesizer employs two
oscillators; one high performance reference crystal oscillator and RF VCO.
The calculation of the PLL output noise requires all individual PLL block noise